1,258 Fake AI Tool Attacks Impersonate Claude and ChatGPT to Hit Developers
Any.Run researchers tracked 1,258 incidents over six months in which threat actors impersonated AI brands to drop infostealers, RATs, and other malware into enterprise networks. Anthropic’s Claude was the most-impersonated brand at 60% of analyses, followed by OpenAI’s ChatGPT and Codex at 33%. The lures arrived through SEO poisoning, malicious ads, fake download pages, and AI-themed artifacts, and the campaigns overwhelmingly relied on ClickFix-style social engineering: 45% of analyses involved a victim being asked to copy and paste a command into a terminal. Fake CLI installers made up 27% of the analyses, traditional ClickFix from fake download pages 17%, and chains starting with abused AI chats or artifacts 25%.
The primary target is developers. Because installing legitimate AI coding tools routinely involves running terminal commands, malicious instructions blend in easily. A single pasted command can expose source code, SSH keys, CI/CD tokens, and cloud or AI API keys. Marketing staff are also hit, with fake ads and poisoned search results aimed at compromising credentials for ad accounts and billing access. Push Security separately reported that AI-themed InstallFix attacks now account for 5% to 10% of its monthly detections and are growing, and flagged a technique called LLMshare in which attackers post malicious content inside shared chats and artifacts on the real AI platforms themselves, making the domain appear trustworthy.
Any.Run recommends organizations publish a short list of approved AI tools with official sources, deploy them through MDM or a software portal, restrict standard users’ ability to paste and run commands, and train employees to verify install instructions against vendor documentation rather than third-party links, ads, or shared chats.
Read the full article
Continue reading at Dark Reading →This is an AI-generated summary. Read the original for the full story.